Anthropic export ban hits Mythos | OpenAI state AG probe | Google AI found liable by Munich

Bottom Line Up Front

The Commerce Department ordered Anthropic to disable Fable 5 and Mythos 5 for foreign nationals on national security grounds, prompting the company to pull both models globally and fly staff to Washington to negotiate. Amazon CEO Andy Jassy set the crackdown in motion by flagging a Mythos jailbreak to senior Trump administration officials, according to people familiar with the matter. A coalition of state attorneys general subpoenaed OpenAI for documents on ChatGPT safety, advertising, engagement and handling of minors as the company moves toward an IPO. The Munich Regional Court found Google liable for false statements in its AI overviews.

I.Top Stories

Trump administration orders Anthropic to disable Fable 5 and Mythos 5 for all foreign nationals

The Commerce Department's Bureau of Industry and Security issued an export control directive Friday cutting off foreign access to both Anthropic models, with a White House letter reaching the company around 5:30 p.m. ET, per Axios. A 1 p.m. ET call from the government had given Anthropic 90 minutes to take Fable and Mythos down on national security grounds, and users lost access to Fable by 10 p.m. Anthropic disabled both models entirely for all customers, saying compliance required cutting off its own foreign-born employees, including U.S.-based staff who are not U.S. citizens. In a public statement, the company said the government had cited only "verbal evidence of a potential narrow, non-universal jailbreak" and argued the same capability is widely available in rival models including OpenAI's GPT-5.5, per Nextgov/FCW.

Read at Axios ↗ • Read at The Washington Post ↗ • Read at Nextgov/FCW ↗

Amazon CEO Andy Jassy triggered the Anthropic crackdown by warning Trump officials about a Mythos jailbreak

Jassy was among the tech leaders who raised national security concerns about Anthropic's Fable 5 and Mythos 5 models in calls to senior Trump administration officials, a person familiar with the matter told Reuters. Information Jassy shared with the administration sparked the sweeping move to halt foreign access to the two models, the Wall Street Journal first reported. The order came just as an earlier dispute between Trump administration officials and Anthropic, which has confidentially filed for a U.S. initial public offering, had shown signs of easing across parts of the government. Jimmy Goodrich, a senior fellow at the University of California's Institute for Global Conflict and Cooperation, said the action also blocks Canadian and British staff at Anthropic from doing research and development.

Read at Investing.com ↗ • Read at WSJ ↗

White House feared a China-linked group had accessed Anthropic's Mythos before the takedown

The White House's decision to impose export restrictions on Mythos was driven in part by fears that a group linked to China had accessed the model, Semafor reported. Chinese government access to Mythos 5 or Fable 5 would carry serious national security risk and could enable distillation, a method in which a smaller model is trained on a more advanced model to replicate its behavior. The White House has not confirmed the report, and an Anthropic spokesperson told Semafor the government did not raise China during its discussions on the export controls. Defense Secretary Pete Hegseth wrote on X on Saturday that "Three months ago, @DeptofWar kicked @AnthropicAI out of our building—forever. Every passing day proves why that was the right move," and Pentagon Chief Information Officer Kirsten Davies separately backed the administration's move in a post on X, saying "Some things are simply more important than revenue cycles, clickbait, and pre-IPO valuation. America First. Always."

Read at Semafor ↗ • Read at The Verge ↗ • Read at X ↗

Multistate attorneys general subpoena OpenAI over ChatGPT safety as the company prepares an IPO

A coalition of state attorneys general served OpenAI with a subpoena seeking documents on a broad range of topics, with New York's attorney general issuing the subpoena on Friday, TechCrunch reported, citing the Wall Street Journal. The subpoena covers the company's advertising, user engagement and retention, model sycophancy, handling of consumer and health data, and treatment of minors and seniors. OpenAI has drawn criticism over ChatGPT allegedly offering encouraging responses to users contemplating suicide or criminal acts, the Associated Press reported. OpenAI said it takes the concerns seriously and will engage constructively with the offices, and noted it had filed confidentially to go public.

Read at TechCrunch ↗ • Read at AP ↗

Munich court preliminarily finds Google liable for false statements in its AI Overviews

The Munich Regional Court preliminarily ruled that Google is liable for false statements generated by its AI Overviews search summaries, requiring the company to prevent dissemination of erroneous claims, Wired reported. The case stems from two publishers who discovered the AI generated summaries linked them, in certain searches, to questionable business practices and subscription related frauds without any basis. The court concluded Google's AI combined data about other companies that had been flagged for possible illicit practices with information about the plaintiffs, producing "independent, new, and substantial statements" that did not appear in the underlying sources. The court rejected Google's defense that user warnings about errors absolved it of liability, finding only Google could modify the technology and was therefore accountable for the outputs.

Read at Wired ↗

II.China Watch

Zhipu releases its most capable open source model and Hong Kong shares jump nearly 48%

Zhipu AI shares briefly hit HK$1,620 in Hong Kong trading on June 15, up 47.68% from the prior close, after the company opened its newest GLM-5.2 model to paid coding plan users, per Caixin. Zhipu described GLM-5.2 as its most capable open source model to date, with a usable 1 million token context window, up from 200,000 in GLM-5.1, released under an MIT license with no regional restrictions. The rollout came two days after Anthropic took its newest models offline under a U.S. government order blocking foreign access. The release lands just ahead of Zhipu's Hong Kong stock lockup expiration, with Chinese AI and semiconductor shares broadly rallying that session. GLM-5.2's permissive licensing, lack of regional restrictions and 1 million token context position it as one of the most accessible substitutes for international developers cut off from U.S. frontier models by new export controls.

Read at Caixin ↗

DeepSeek's V4 model was co-designed with Huawei's newest AI accelerator, cutting inference costs by 75%

A SemiAnalysis report concluded that DeepSeek's V4 model and Huawei's Ascend 950DT AI accelerator were engineered around each other from the ground up, Pandaily reported. The pairing let DeepSeek slash its V4-Pro API price by 75%, with the company saying the new tier remains profitable. Huawei has pulled forward volume deployment of the 950DT in its cloud to August 2026, with ByteDance reportedly reserving roughly half of initial production and Alibaba and Tencent placing orders in the tens of thousands. China Mobile is bringing online 776 nodes built on the new chip, the report said. The shift extends the domestic training stack into commercial inference at scale, narrowing the practical bite of U.S. Commerce Department export controls on advanced AI accelerators sold to China.

Read at Pandaily ↗

Beijing formally defines "AI agent" as the gaokao becomes a proving ground

Three Chinese ministries jointly issued the country's first formal definition of "AI agent" as a regulated product category, Pandaily reported, recognizing the technology for the purposes of registration and oversight. The definition arrived as Alibaba, Tencent and Baidu rolled out free AI agents to guide the 12.9 million students sitting China's national college entrance exam, known as the gaokao. Alibaba's Tongyi Qianwen agent generated more than 13 million guidance reports during the 2025 gaokao season, per the report. Tencent paired its Yuanbao agent with mobile consulting vans in three cities, and Baidu's ERNIE Bot agent drew on 20,000 university seniors across 2,200 institutions. China had registered 868 generative AI services and 530 generative AI apps under the Cyberspace Administration of China's filing regime as of April. Formalizing the category gives Beijing a concrete regulatory hook for content review, security assessments and accountability rules as agentic systems begin operating across consumer markets at scale.

Read at Pandaily ↗

Meta moves to unwind $2 billion Manus acquisition under Beijing's divestiture order

Meta has begun dismantling its $2 billion acquisition of Manus and completed an operational separation from the Chinese-founded AI startup, halting data sharing between the two companies, per TechCrunch. Beijing issued the divestiture order roughly two months ago on national security grounds, citing potential violations of technology export controls and foreign investment rules. The Manus co-founders are in preliminary talks to raise about $1 billion from outside investors to reclaim the startup, a path that could lead to a Chinese joint-venture structure and an eventual Hong Kong listing. Manus investors including Tencent, HSG and ZhenFund have indicated they will cooperate with the unwinding, while California-based Benchmark has already received its proceeds from the acquisition.

Read at TechCrunch ↗

III.Federal Policy Tracker

Senate Judiciary will mark up the NO FAKES Act on June 18

The Senate Judiciary Committee will hold a business meeting at 10:15 a.m. ET on June 18 to consider S.4591, the NO FAKES Act, per Congress.gov. The same agenda includes S.1133 on federal court media coverage, S.1146 on televising Supreme Court proceedings, several judicial nominations, and the nomination of Don Richard Berthiaume, Jr. as Department of Justice Inspector General. The NO FAKES bill's title states it would "protect intellectual property rights in the voice and visual likeness of individuals."

Read at Congress.gov ↗ • Read at Senate Judiciary ↗

VA Inspector General faults Veterans Health rollout of generative AI chat tools without proper governance

The Department of Veterans Affairs provided clinical staff with generative AI chat tools, including VA GPT and Copilot Chat, without adequate oversight or safeguards, the VA Office of Inspector General concluded in a report released Thursday, per FedScoop. The OIG said the lack of governance created risks for patient safety and limited the department's ability to monitor errors during its October 2025 to February 2026 study period. The watchdog said efforts to reduce "unnecessary levels of bureaucratic oversight" under the Office of Management and Budget's 2025 AI memo "may have introduced more vulnerabilities," with VA AI leaders prioritizing rapid access over safeguards deemed impractical. VA concurred with the OIG recommendations to address AI chat tool governance, evaluate the tools as high impact, and integrate AI risk monitoring into existing patient safety programs.

Read at FedScoop ↗

FBI dismantles Chinese AI powered phishing service Outsider Enterprise with Google and Black Lotus Labs

The FBI, working with Google and Lumen's Black Lotus Labs, has dismantled a Chinese AI powered phishing-as-a-service operation called Outsider Enterprise that ran roughly 9,000 fake websites and more than 1 million fraudulent URLs since at least 2023, per BleepingComputer. Authorities estimate the campaigns led to the theft of more than 3.8 million credit card records and about $1.9 billion in losses. The technical takedown, part of the FBI's larger Operation Riptide, seized multiple administration servers, a Shopify storefront, a test account used by the threat actor, and about $100,000 in USDT (Tether stablecoin) from the operation's payment wallets. Google filed a civil suit targeting the operation's infrastructure and is coordinating with AT&T, T-Mobile and Verizon to block fraudulent text messages before they reach subscribers.

Read at BleepingComputer ↗

IV.Industry & Market Watch

KPMG pulls "Redefining excellence in the age of agentic AI" report after GPTZero flags AI hallucinations

KPMG has pulled its "Redefining excellence in the age of agentic AI" report, published in October 2025, after multiple organizations told the Financial Times that claims about their AI usage in the report were untrue, per TechCrunch. Research group GPTZero identified the inaccuracies and concluded they stemmed from AI hallucinations (fabricated content the model presented as fact), suggesting the firm had used AI to help write the report. UBS, the UK's National Health Service, Swiss Federal Railways and Transport for London all told the FT the report's claims about their AI usage were either untrue or misleading. A KPMG spokesperson said the firm removed the report while conducting its own investigation, and said staff are required to apply human oversight to validate AI content.

Read at TechCrunch ↗

The AI layoff wave hits 974 tech job cuts per day as Andreessen calls AI the "silver bullet excuse"

Tech companies have laid off nearly 150,000 people so far this year through about 363 separate layoff events, a pace of roughly 974 workers per day and 44% faster than 2025, per TrueUp data cited by TechCrunch. Last month was the highest single month of tech layoffs in two years at nearly 40,000 cuts, with AI cited as the leading reason for layoffs across every industry for the third month running, per outplacement firm Challenger, Gray & Christmas. Venture capitalist Marc Andreessen told podcaster Harry Stebbings that every large company is overstaffed by at least 25% and called AI "the silver bullet excuse" for layoffs often driven by mismanagement. Block CEO Jack Dorsey initially blamed AI tools after nearly halving the company earlier this year, before acknowledging that Block had over-hired during the pandemic.

Read at TechCrunch ↗

V.Global & Geopolitics

Canadian Prime Minister Mark Carney says U.S. AI export ban shows risk of relying on big AI models

Canadian Prime Minister Mark Carney said on Sunday that the U.S. export ban on Anthropic's latest AI models illustrates the risk of "over-reliance on certain models," speaking to reporters during a visit to Ireland, per the Anadolu Agency. Drawing a parallel with the 2008 global financial crisis, Carney said the AI sector faces similar systemic vulnerabilities and called for countries and companies to seek "redundancy and diversity" in AI development. He linked the U.S. AI curbs to Canada's broader push to diversify trade and technology, with more than 70% of Canadian exports going to the U.S. and Carney having set a goal of doubling non-U.S. exports over the next decade, per AP. Carney does not have a bilateral meeting with President Trump scheduled at the G7 summit in France this week.

Read at Anadolu ↗ • Read at AP ↗ • Read at Bloomberg ↗

Kazakhstan signs $10 billion AI infrastructure accords with Firebird and Nvidia support

Kazakhstan signed accords on Monday with Nvidia backed startup Firebird Inc. on AI computing projects that could bring as much as $10 billion in investment, the country's AI and digital development ministry said, per Moneycontrol. The accords include a strategic cooperation agreement on developing AI infrastructure and terms for a planned "Data Center Valley" project in northeastern Kazakhstan, set to start in 2027 with a $5 billion phase one investment. Nvidia is supporting Firebird with project design and GPUs, the ministry said.

Read at Moneycontrol ↗ • Read at Bloomberg ↗